John1231983

I meant I want to download more data from google, instead of small number of user. Is it possible

@naiiytom the dataset is not good. Do not use it

@nidetaoge : I also have same problem when add other loss. The D loss goes to zero and the generator loss goes up. How do you solve the issue?

@ericj974 : Thanks for your code. Could you tell me the performance of ResnetXt in comparison with Resnet 101 and Resnet 50?

@ericj974 : What is your performance of resnetxt ? How do you train from scratch? I tried the command `python3 coco.py train --dataset=/path/to/coco/ ` (deleted `--model=coco`) but it got error...

@Cprune: your mAP for bbox is better than the last version of repo (0.37 / 0.34) for resnet 101. Could you also show the mAP for segmentation? The repo achieved...

Good job @Cpruce : Could you please share your pre-train coco using resnet101? Can I use it as pretrain for another dataset for instance segmentation I am very interesting why...

Sorry. I means resnet50(second image). Have you give me your pretrain resnet 50 on coco ?

@Cpruce : So, you just use the matterport pretrain model and it give 0.37 AP. Am I right? Because the report of matterport shows 0.34

Yes. I have 1 titanx pascal 12gb. So if I understand correctly, you start to train coco with imagenet resnet50 pretrained. You also finetune config. After training, you got a...